Bug #3542: cssens bugs - ctools - CTA IRAP Project Gateway

Bug #3542

cssens bugs

Added by Sadeh Iftach almost 4 years ago. Updated almost 3 years ago.

Status:

Closed

Start date:

02/17/2021

Priority:

Normal

Due date:

Assigned To:

% Done:

100%

Category:

Target version:

2.0.0

Duration:

Description

Hello,

Here are a couple of issues I’ve run into with cssens

too-low values:

In https://github.com/ctools/ctools/blob/devel/cscripts/cssens.py#L387
I’ve run into errors of the kind:

              File "cssens.py", line 754, in run
                result = self._e_bin(ieng)
              File "cssens.py", line 693, in _e_bin
                result = self._get_sensitivity(emin, emax, self._models)
              File "cssens.py", line 401, in _get_sensitivity
                prefactor.value(crab_prefactor * test_crab_flux)
              File "/afs/ifh.de/group/cta/scratch/sadeh/software/anaconda/Anaconda3-2020.11-Linux-x86_64/envs/ctools_stable/lib/python3.7/site-packages/gammalib/opt.py", line 459, in value
                return _opt.GOptimizerPar_value(self, *args)
            ValueError: *** ERROR in GOptimizerPar::factor_value(double&): Invalid argument. Specified value factor 9.99999999999999e-11 is smaller than the minimum boundary 1e-10.

I suggest something like:

            min_pref = prefactor.min()
            val_now = crab_prefactor * test_crab_flux
            val_now = max(val_now, min_pref * 1.01)
            prefactor.value(val_now)

division by zero

In https://github.com/ctools/ctools/blob/devel/cscripts/cssens.py#L597
I’ve encountered division by zero errors.

I suggest something like:

        rxy_norm = (mean_xx - mean_x * mean_x) * (mean_yy - mean_y * mean_y)
        if rxy_norm < 1e-10:
            rxy_norm = 1
        else:
            rxy_norm = math.sqrt(rxy_norm)

        rxy = (mean_xy - mean_x * mean_y) / rxy_norm

Recurrence

No recurrence.

History

#1 Updated by Sadeh Iftach almost 4 years ago

These extremely low values for the prefactor correspond eventually to completely unrealistic sensitivities:

        loge   emin   emax  crab_flux   photon_flux   energy_flux   sensitivity  regcoeff  nevents      npred
0 -0.200171  0.501  0.794   0.000454  1.212797e-14  1.204168e-14  2.609276e-14   0.23334     22.0  21.999989

where the calculation was done for a 10 second exposure!

This seems related to using a background model based on:

<source_library title="source library">
  <source name="CTABackgroundModel" type="CTAIrfBackground" instrument="CTA">
    <spectrum type="PowerLaw">
      <parameter name="Prefactor" value="1" error="0" scale="1" min="0.001" max="1000" free="1" />
      <parameter name="Index" value="0" error="0" scale="1" min="-5" max="0" free="1" />
      <parameter name="PivotEnergy" value="1" scale="1000000" min="0.01" max="1000" free="0" />
    </spectrum>
  </source>
  <source name="source_00047513" type="PointSource" tscalc="1">
    <spectrum type="PowerLaw">
      <parameter name="Prefactor" value="1" error="0" scale="1e-16" min="1e-10" max="10000000000" free="1" />
      <parameter name="Index" value="-2.5" scale="1" min="-5" max="0" free="0" />
      <parameter name="PivotEnergy" value="1" scale="1000000" min="0.001" max="1000" free="0" />
    </spectrum>
    <spatialModel type="PointSource">
      <parameter name="RA" value="265.97" scale="1" min="-360" max="360" free="0" />
      <parameter name="DEC" value="-29.38" scale="1" min="-90" max="90" free="0" />
    </spatialModel>
  </source>
  <source name="merged_mapcube_models" type="DiffuseSource">
    <spectrum type="PowerLaw">
      <parameter name="Prefactor" value="1" scale="1" min="1e-09" max="1000000000" free="1" />
      <parameter name="Index" value="0" scale="1" min="-5" max="0" free="1" />
      <parameter name="PivotEnergy" value="1" scale="1000000" min="0.1" max="10" free="0" />
    </spectrum>
    <spatialModel type="DiffuseMapCube" file="output/sense_0/source_00047513/source_00047513_pl_0_merged_mapcube_models.fits">
      <parameter name="Normalization" value="1" scale="1" min="1e-09" max="1000000000" free="0" />
    </spatialModel>
  </source>
</source_library>

and includes a DiffuseMapCube

I’ve tried using the same model, but not allowing the merged_mapcube_models background parameters to be fit. (Only CTABackgroundModel and the test source, source_00047513, have free parameters.) In this case, I get what looks like the expected answer, of ~10^-10 sensitivity for 10sec:

       loge   emin   emax  crab_flux   photon_flux   energy_flux  \
0 -0.200171  0.501  0.794   1.812952  4.845686e-11  4.811212e-11
1 -0.000077  0.794  1.259   2.237868  3.028069e-11  4.765867e-11
2  0.199984  1.259  1.995   2.639513  1.803496e-11  4.499607e-11

    sensitivity  regcoeff  nevents      npred
0  1.042527e-10  0.970945     35.0  30.346305
1  1.031549e-10  0.946488     26.0  19.374530
2  9.753276e-11  0.978457     20.0  14.034580

So wither something is be going on with the DiffuseMapCube, or the inclusion of multiple background models on it’s own results in too much complexity for the fit.

#2 Updated by Knödlseder Jürgen over 3 years ago

Target version set to 2.0.0

Iftach, if you want to propose some code to get merged in ctools please go ahead with coding and testing and making a pull request.

#3 Updated by Knödlseder Jürgen over 3 years ago

As far as I can see, the corresponding code has now be integrated. Iftach, can you confirm that this issue is done?

#4 Updated by Sadeh Iftach over 3 years ago

Status changed from New to Resolved
% Done changed from 0 to 100

#5 Updated by Knödlseder Jürgen almost 3 years ago

Status changed from Resolved to Closed

Also available in: Atom PDF

ctools

Issues

Tags